Separate understanding from transformation.
The foundation reconstructs verifiable evidence from source. Target modernization stages are designed to reason within that context, propose bounded changes, and verify them independently.
Current deterministic foundation
Source repositoryJavaScript and TypeScript are the current analysis focus. This repository layout is schematic.
Repository intelligence
VyrixScout inventories files and reconstructs modules and dependencies. Discovery establishes the system boundary before deeper analysis.
classifyScoreRepresentative function name from the public sample; this AST is an explanatory schematic.
Structural understanding
NolaraStruct models syntax, scopes, and symbols. Stable structure makes program relationships inspectable instead of leaving them buried in text.
Function entryEntry into classifyScore. This is static control flow, not an execution trace.
Control & data flow
Control flow exposes possible paths. Data flow concerns how values relate across program operations. The public sample demonstrates selected intraprocedural CFG and reachability; DFG belongs to the broader documented foundation.
scoreparameter>= 70comparison- Branch condition
Illustrative value dependency, not captured DFG output. The public CFG sample excludes data-flow analysis.
Source revisionArtifacts need source provenance to remain interpretable.
Deterministic evidence
Canonical artifacts retain relationships and source provenance. Reproducibility lets engineers compare and revisit a model; it does not guarantee a complete model or behavioral equivalence.
Future modernization extension
Bounded AI reasoning
Explain architecture, identify risks, and propose strategies through a governed AI gateway. Source evidence remains the authority for the system model.
ZelvoxForge
Propose governed transformations and target code generation (Rust, Axum, SQLx, and modern TypeScript) with defined inputs, outputs, and review boundaries.
VymosGate
Independently assess migration artifacts through isolated compiler sandboxes, differential execution testing, and human approval gates.
Conceptual target flow. No autonomous migration service is offered by this website.
- 01
Evidence
Versioned source context
- 02
Bounded AI
Typed inputs & limited tools
- 03
Change plan
Explicit transformation scope
- 04
Candidate
Proposed target-specific code
- 05
Verification
Independent checks & results
- 06
Human review
Approve, revise, or reject
AI proposes. Independent checks supply evidence. People decide. These future controls require implementation and evaluation.
The bounded AI reasoning model
The central architectural thesis of LegacyExodus is that generative AI must be bounded by deterministic evidence:
Deterministic evidence defines what is known.
AI helps reason about what comes next.
Independent verification determines what can be accepted.
Rather than granting models unconstrained read/write access to codebases, our planned architecture enforces six non-negotiable boundaries:
- Versioned Context: The AI gateway receives immutable, source-grounded graph snapshots produced by NolaraStruct, never speculative or unanchored text prompts.
- Structured Inputs & Outputs: All reasoning interchanges use strictly typed JSON schemas with bounded token windows rather than conversational chat.
- Explicit Evidence References: Every suggested refactor, transformation contract, or risk hypothesis must cite the underlying AST node, CFG edge, or symbol identifier.
- Constrained Tool Permissions: Models operate with strictly scoped, read-only graph query tools and isolated execution sandboxes without open network access.
- Checkpoints & Gates: Candidate transformations must clear automated structural invariants before reaching compilation or differential test stages.
- Human Oversight Authority: The enterprise architect holds final release authority. AI proposes; independent verification tests; engineers approve.
Extend the language boundary carefully.
Source adapters must preserve language-specific meaning. A shared intermediate representation (IR) is useful only where semantic mappings and transformation contracts can be defended without loss.
JavaScript / TypeScript
Repository discovery & semantic analysis
No shipped adapters or committed sequence.
Meaning across languages.
Language-aware adapters, structured knowledge & IR, explicit mappings and transformation contracts.
Bounded transformation
Target-specific generation and independent verification.
Rust is an intended target. Other destinations require their own semantic mappings and evidence.
The trust boundaries
Source to evidence
Analysis derives relationships from source rather than generated descriptions. Provenance and known static-analysis limitations are necessary context for interpreting any artifact.
Evidence to proposals
The intended AI boundary permits reasoning over a bounded system model. A proposal is not verified behavior. Unified IR and SSA are target representations, not capabilities this website presents as complete.
Proposals to acceptance
Independent verification, isolated compiler sandboxes, repair loops, and human migration approval are planned controls. The intended fail-closed policy is to withhold acceptance when required evidence or checks are missing.
These future boundaries describe architecture intent. Their implementation and end-to-end effectiveness remain to be established.
Reproducibility has a job
Stable artifacts allow engineering teams to revisit a model and compare results. A repeatable result can still be incomplete. Determinism, reconstruction fidelity, and behavioral equivalence are distinct engineering properties that must be evaluated individually.
Start with understanding.
Follow the engineering, explore the approach, or get in touch.